Ecosystem scale training improves agent skills in shared tool platform
MCPDisCo: Seesaw-Mitigated Tuning for MCP-Native Agents
Artificial Intelligence
Summary
Many digital services offer different tools that a smart program (agent) can use by following common rules called the Model Context Protocol. Teaching an agent to use these tools well requires learning to think and act across many services and tools. The authors created MCPDisCo, a training method that helps agents think better and choose tools more effectively by reducing conflicts between thinking and doing. This approach lets agents perform well on tasks involving many tools, even those they haven't seen before.
What this means in practice
- •For software engineers: Train intelligent agents to effectively use numerous heterogeneous tools across web services with improved reasoning and execution.
- •For cloud platform developers: Enhance server ecosystems by enabling agents trained with MCPDisCo to interact with diverse tools reliably and adapt to new environments.
Authors
Kounianhua Du, Jianxing Liu, Shuai Shao, Ke Li, Xing Sun, Weinan Zhang
Abstract
The Model Context Protocol provides a shared interaction substrate for the agentic web, allowing heterogeneous services to expose capabilities through a standardized interface. Learning to use this ecosystem entails a hierarchy of capabilities: tool-format skills, server-level transfer, and cross-server composition. We present MCPDisCo, an ecosystem-scale reinforcement learning framework for MCP-native agents. Using a Thought-Markov Decision Process (Thought-MDP), the framework explicitly represents internal reasoning alongside routing, execution, and termination and trains on real-server, long-horizon interactions spanning 70 MCP servers and 527 tools across six domains. We study the reasoning--tool seesaw, a joint-learning failure mode in which improvement in one objective can accompany deterioration in the other. To mitigate potential interference, MCPDisCo combines token-masked reward attribution with bidirectional gradient surgery: the former separates direct supervision for reasoning and action spans, while the latter removes conflicting components from both objective gradients in shared parameters. MCPDisCo achieves higher success with fewer chain-of-thought tokens. On the out-of-distribution MCP-Atlas and the GAIA benchmark, MCPDisCo maintains monotone gains, confirming that the learned behavior generalizes beyond the training server distribution and broader agentic tasks. Model weight and implementation are available.